Presentation | 2015-03-16 A Proposal of Novel Data Detection Method and Its Application to Incremental Learning for RBMs Masahiko OSAWA, Masafumi HAGIWARA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Incremental learnings without destruction of the existing memory are often difficult for deep learning, since most of the weights change. In this paper, we propose an incremental learning method for Restricted Boltzmann Machines (RBMs). First, we suggest that trained RBMs can detect novel data using their energy function. Second, we propose the incremental learning method for these novel data by adding some new units and combine them with the trained network. The proposed method can memorize novel data and without degradating learned contents. According to the evaluation experiments, it is suggested that RBMs can find novel data using the energy function and learn without destruction of learned contents. Furthermore, in the task of the tick-tack-toe, the agents using the proposed method could get strategies progressively and they were better than agents without these methods even if the latter learned much more data. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Restricted Boltzmann Machine / Deep Learning / Associative Memory / Incremental Learning |
Paper # | MBE2014-167,NC2014-118 |
Date of Issue |
Conference Information | |
Committee | NC |
---|---|
Conference Date | 2015/3/9(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Neurocomputing (NC) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A Proposal of Novel Data Detection Method and Its Application to Incremental Learning for RBMs |
Sub Title (in English) | |
Keyword(1) | Restricted Boltzmann Machine |
Keyword(2) | Deep Learning |
Keyword(3) | Associative Memory |
Keyword(4) | Incremental Learning |
1st Author's Name | Masahiko OSAWA |
1st Author's Affiliation | Faculty of Science and Technology, Keio University() |
2nd Author's Name | Masafumi HAGIWARA |
2nd Author's Affiliation | Faculty of Science and Technology, Keio University |
Date | 2015-03-16 |
Paper # | MBE2014-167,NC2014-118 |
Volume (vol) | vol.114 |
Number (no) | 515 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |